ESP32 S3 Box3

Mine certainly need calibration - 22.9 degrees celsius, >3 degrees difference with actual temperature. Could be due to the wrong sensor being used? AHT10?

Also the update frequency seems too high

[16:58:19][D][aht10:097]: AHT10 is busy, waiting...
[16:58:19][D][sensor:094]: 'Temperature': Sending state 22.90573 °C with 2 decimals of accuracy
[16:58:19][D][sensor:094]: 'Humidity': Sending state 42.21153 % with 2 decimals of accuracy
[16:58:19][W][component:214]: Component aht10.sensor took a long time for an operation (0.07 s).
[16:58:19][W][component:215]: Components should block for at most 20-30ms.

Supposedly the device should also be able to act as a bluetooth proxy; is this realistic though to run in parallel with the voice assistant?

As far as I know, Bluetooth proxy is using a lot of the resources from such devices.
So I think, it will not work very stable / reliable… at that point.

MAYBE, there are ways to get over such limitations somehow, but I don’t think, that this is in the focus for the team.

This link: https://github.com/esphome/feature-requests/issues/2475 seems to imply that the radar can be integrated into esphome (given that the request was closed?)

only the issue reported in the “firmware” repo was closed, because the current firmware is “generally working”.

The issue / feature request for the radar is still open… :wink:

BTW:
It seems, that there is also an issue with the i2c component in combination with ADF … which is stopping VA to work under certain conditions.

Therefore, I have removed the i2c and temperature sensor again, until this issue might be fixed.

This issue with the i2c will probably also affect the radar, because it’s also using the i2c bus, afaik

I see what you mean. Thank you for the clarification.

Can you please explain how to use ollama assistant for text to speech? I use ollama with llama2_13B for HA’s chat box assistant but I would like to use it for speech too instead of the Google Home.

The request is open.

Is it just my unit or the ok nabu wakeword doesn’t works most of the time (almost never ?) and the speaker is almost not hearable ?

For me wakeword works but the speaker volume is also very low. I hardly can hear a response even in a quiet room. So no matter how it performs otherwise, it is not very useful as a voice assist right now.

3 Likes

any chance to buy ESP32 S3 Box3 from somewhere without waiting months?

Yeah I dunno how something that is quiet to the point of uselessness became recommended.

Can we plug into a powered speaker?

1 Like

The speaker is indeed much much softer than e.g. the atom echo. My created wake words work oke, but the BOX-3 is harder to reach and is to low in speaker volume.

Its the software. Out of the box it includes an MP3 player which was ok volume wise (not hi-fi but okayish)

That should be fixable then!

That’s good news. I created a new issue for this problem on Github/ESPHome here:
https://github.com/esphome/firmware/issues/139

Well when I say fixable, finding and understanding the software is not easy. The integration for esp_adf is still in a PR.

1 Like

Volume on mine is quite acceptable for me.
For comparison, I did an audio dB measurement at 50cm using the Alan TTS model in piper. Result at 50cm was 71dB max.

I have the radar working. It was quite simple:

binary_sensor:
  - platform: gpio
    pin:
      number: GPIO21
    name: "Radar detect"
    disabled_by_default: true
    device_class: "occupancy"
    entity_category: diagnostic

Note, this is only using the default detection settings of the chip. The I2C bus would need to be used to adjust to any desired settings, but note that the use of the I2C bus at this point seems to conflict with the microphone and general VA usage.

As it is, it seems to detect as follows:
ON: immediately with any movement
OFF: 1s after on and waits for further movement

The range seems to be 0-5 metres
Horizontal cover at 0.5m is about 160 deg
Vertical cover at 0.5m is about 100 deg
Sensitivity seems to be movements about 3-5cm at any distance

Adding the following filter produces a more stable output that doesn’t go on and off all the time:

    filters:
      - delayed_off: 10s
3 Likes

Did you make any tweaks to the yaml as shared by the esphome/ha team?

Here is an updated RADAR config that:

  • adds an occupancy sensor to the Sensors section.
  • adds a configurable occupancy duration to the Config section (default: 60s).
  • adds a switch to the Config section that mutes the box and turns off the backlight when nobody is present (default: not enabled).

The last helps if on battery and also reduces network chatter.
The first can be used in HA automations to turn on lights, fans, etc.

switch:
  - platform: template
    name: "Mute when absent"
    id: mute_when_absent
    icon: mdi:account-right-arrow
    optimistic: true
    entity_category: config
    restore_mode: RESTORE_DEFAULT_OFF

number:
  - platform: template
    name: "Presence duration"
    id: radar_delayed_off
    icon: mdi:account-clock
    optimistic: true
    restore_value: true
    initial_value: 60
    min_value: 0
    step: 5
    max_value: 300
    unit_of_measurement: s
    entity_category: config
    mode: box

binary_sensor:
  - platform: gpio
    pin:
      number: GPIO21
    name: "Presence detect"
    disabled_by_default: false
    device_class: "occupancy"
    filters:
      - delayed_off: !lambda return id(radar_delayed_off).state * 1000;
    on_release:
      then:
        - if:
            condition: 
              switch.is_on: mute_when_absent
            then:
              - switch.turn_on: mute
              - light.turn_off: led
    on_press:
      then:
        - if:
            condition: 
              switch.is_on: mute_when_absent
            then:
              - switch.turn_off: mute
              - light.turn_on: led

It is working well for me. I’d be interested to hear how well it works for others too. There was an initial issue for me with the device starting muted and not waking due to the absence switch defaulting to off, however I haven’t been able to replicate this.

Remember to clean your build files before install.

Enjoy.

3 Likes